Study of the Short-Term Generation Block Frequency Forecasting of a Power Generation Utility Using Artificial Neural Network

Abstract

Frequency forecasting a very vital aspect in power system operation and control. For In traditional approaches, frequency prediction algorithms prepared based on pattern matching of the previous frequencies, but here in this paper an algorithms is prepared which will include the various other non linear parameters also such as actual power generated, load demand in term of scheduled power generation allocated to particular utility. This deficit/surplus power is taken as one of the key factor for forecasting the frequency to a fair degree of accuracy. ANN's ability to provide most accurate forecasting results with smaller processing time is utilized here in this paper.In this work, we will study short-term frequency prediction for a thermal power generating station station with the help of ANN. We will utilize the past two year data of plant's suggested generation, declared capacity and frequency for each block of 15 minutes to train the Artificial Neural Network. Keywords: ABT, Unscheduled Interchange, NRLDC Artificial Neural Network, Load Dispatch --------------------------------------------------------------------***---------------------------------------------------------------------- BACKGROUNDFrequency forecasting in power system is quite tedious job as nature of frequency is stochastic. It varies with declared capacity of the station, grid network frequency demand, surplus/deficit power in system. Besides this, it also depends on frequency value in previous blocks. Considering these all-crucial parameter numerous forecasting techniques evolved [2], [3] considering stochasticity of the frequency.In conventional methods, such as time series, regression theory model, Kalman Filter and Spectral Expansion technique considered frequency values of the previous times with known per day or per session basis periodicities. Authentic frequency forecasting is not viable with these methods, as they do not adequately reassemble non-linearity between the frequency and various other parameters associated with it.Apart from this, these approaches demonstrate larger processing time, sluggish performance and incorrect forecasting due to rapidly season change. Artificial Intelligence (AI) based approaches such as ANN, Genetic Algorithams, Experts Systems and fuzzy logic have shown their excellence over the conventional methods of forecasting. The Artificial Neural Network learns from experience and easily embodied to adopt the non-linear relationship between frequency and other factors associated with historical data. Hence, with application of the ABT for enforcing charges on power drawl that is sensitive to the system frequency, power utility operator focused on designing predicting generation block frequency for earning UI charges and scheduling of the generators accordingly.